Visual Inference for Modeling and Reasoning with Bayesian Networks

Failure Modes and Effects Analysis (FMEA) has been the de facto standard, in designing and analyzing components or systems mainly to manage potential risks, for many years. In recent years many endeavours were made to move from classical FMEA which focuses on qualitative evaluation to more quantitative assessments. One idea is to leverage Bayesian networks to model function and failure modes according to certain probabilities. Research regarding visualization of Bayesian networks however stayed consistently low over time. In this work we want to propose different visualization and interaction elements for Bayesian networks that are either common in other fields such as direct data manipulation or novel approaches like visualization of cause-and-effect relationships in a node-link graph. A qualitative study with three participants who conducted a one-hour interview using the implemented tool was evaluated to determine the usability of the proposed visualization approaches. The achieved System Usability Scale (SUS) score as well as our questionnaire about the usability of the implemented features both suggest a good general usability even though viability for more complex models was rated slightly below average. We suggest many ideas for possible future research as we think there is a potential for improvement not only building upon this work but in the research of visualization in Bayesian networks in general.